diff --git a/quad/src/quad_app/sensor_processing.c b/quad/src/quad_app/sensor_processing.c index ec33504f517e5e27fc0db34a36ab939fab9bd5a5..e91da27f5910fa10ad241d64f6f904735aaa9806 100644 --- a/quad/src/quad_app/sensor_processing.c +++ b/quad/src/quad_app/sensor_processing.c @@ -26,6 +26,7 @@ #define MAX_VALID_LIDAR (10.0) // Maximum valid distance to read from LiDAR to update int sensor_processing_init(sensor_t* sensor_struct) { + // 10Hz cutoff at 200hz sampling float a0 = 0.0200833310260; float a1 = 0.0401666620520; float a2 = 0.0200833310260; @@ -35,6 +36,11 @@ int sensor_processing_init(sensor_t* sensor_struct) { sensor_struct->accel_y_filt = filter_make_state(a0, a1, a2, b1, b2); sensor_struct->accel_z_filt = filter_make_state(a0, a1, a2, b1, b2); + // 10Hz filters for bias-corrected euler rates + sensor_struct->phi_dot_filt = filter_make_state(a0, a1, a2, b1, b2); + sensor_struct->theta_dot_filt = filter_make_state(a0, a1, a2, b1, b2); + sensor_struct->psi_dot_filt = filter_make_state(a0, a1, a2, b1, b2); + //1 Hert filter float vel_a0 = 2.3921e-4; float vel_a1 = 4.7841e-4; @@ -56,19 +62,34 @@ int sensor_processing_init(sensor_t* sensor_struct) { } - // Focal length in mm = 16 +// Focal length in mm = 16, in pixels is below // static float focal_length_px = 16.0 / (4.0f * 6.0f) * 1000.0f; //original focal lenght: 12mm pixelsize: 6um, binning 4 enabled /* - * Modifies the pixel flow values to account for rotations + * Convert integral frame flow in radians to velocity in m/s * Theta = pitch, phi = roll + * Note that we pass phi and theta as angles, but psi dot, because we don't have a complementary-filtered psi (yaw) */ - void flow_gyro_compensation(px4flow_t* flow_data, double distance, double phi, double theta, double psi_d) { + void flow_gyro_compensation(sensor_t* sensor_struct, double distance, + double phi, double theta, double psi_d_new) { + //------ Gyro compensation stuff. It seems to make the quadcopter unstable, although all the math checks out and data seems better + /* + // The reason for converting back to euler rates instead of using raw gyroscope data is so that we can use + // The complementary-filtered angles, which will prevent gyroscope drift from creating position drift static double last_phi = 0; static double last_theta = 0; // Calculate difference in angles - double phi_d = phi - last_phi; - double theta_d = theta - last_theta; + double phi_d_new, theta_d_new; + float loop_dt = get_last_loop_time(); + if (loop_dt != 0) { // divide by zero check + phi_d_new = (phi - last_phi) / loop_dt; + theta_d_new = (theta - last_theta) / loop_dt; + } else {phi_d_new = theta_d_new = 0;} + + // Run low-pass filters on euler angle rates + float phi_d = biquad_execute(&sensor_struct->phi_dot_filt, phi_d_new); + float theta_d = biquad_execute(&sensor_struct->theta_dot_filt, theta_d_new); + float psi_d = biquad_execute(&sensor_struct->psi_dot_filt, psi_d_new); // Convert angles to body rotations (gyroscope equivalents) ///////////////////------- Inverse of AEB matrix -------////////////////// @@ -80,33 +101,45 @@ int sensor_processing_init(sensor_t* sensor_struct) { double cos_theta = cos(theta); double cos_phi = cos(phi); + // We re-calculate p, q, r instead of using the gyroscope values because these are calculated using + // the complementary filter pitch and roll, which eliminates drift over time double p = phi_d - sin_phi*psi_d; double q = cos_phi*theta_d + cos_theta*sin_phi*psi_d; double r = -sin_phi*theta_d + cos_phi*cos_theta*psi_d; + */ - // TODO: Clamp correction to maximum search radius - // TODO: Verify p vs q here - // Compensates for angle rotation using the gyroscope - double flow_x_rad_corr = flow_data->flow_x_rad - p; - double flow_y_rad_corr = flow_data->flow_y_rad - q; + // Convert rotations to rotation rates + double flow_x_rad_rate = sensor_struct->optical_flow.flow_x_rad / sensor_struct->optical_flow.dt; + double flow_y_rad_rate = sensor_struct->optical_flow.flow_y_rad / sensor_struct->optical_flow.dt; + + // Add p to flow_x_rad_rate to add gyro compensation (Currently disabled) + // Add q to flow_y_rad_rate + double x_rad_rate_corr = flow_x_rad_rate;// + p; + double y_rad_rate_corr = flow_y_rad_rate;// + q; - // Swap x and y to switch from rotation around an axis to movement along an axis - double x_dist = -flow_y_rad_corr * distance; - double y_dist = flow_x_rad_corr * distance; - if (flow_data->quality > PX4FLOW_QUAL_MIN) { - flow_data->xVel = x_dist / flow_data->dt; - flow_data->yVel = y_dist / flow_data->dt; + // Only accumulate if the quality is good + if (sensor_struct->optical_flow.quality > PX4FLOW_QUAL_MIN) { + // Swap x and y to switch from rotation around an axis to movement along an axis + // Y is negative because some reason? + // We simply multiply by distance, because for small angles tan(theta) = theta. + // Also, the internal PX4Flow code works under the small angle assumption, + // so not doing trig here makes it more accurate than doing trig + sensor_struct->optical_flow.xVel = -y_rad_rate_corr * distance; + sensor_struct->optical_flow.yVel = x_rad_rate_corr * distance; } // Gradually decay towards 0 if quality is bad else { - flow_data->xVel *= PX4FLOW_VEL_DECAY; - flow_data->yVel *= PX4FLOW_VEL_DECAY; + sensor_struct->optical_flow.xVel *= PX4FLOW_VEL_DECAY; + sensor_struct->optical_flow.yVel *= PX4FLOW_VEL_DECAY; } - // Store angles for next call + // Un-comment if using gyroscope compensation + /* + // Store angles for next time last_phi = phi; last_theta = theta; + */ } @@ -188,29 +221,35 @@ int sensor_processing(log_t* log_struct, user_input_t *user_input_struct, raw_se sensor_struct->gyr_y = raw_sensor_struct->gam.gyro_yVel_q; sensor_struct->gyr_z = raw_sensor_struct->gam.gyro_zVel_r; + double loop_dt = get_last_loop_time(); // Complementary Filter Calculations - sensor_struct->pitch_angle_filtered = ALPHA * (sensor_struct->pitch_angle_filtered + sensor_struct->theta_dot * get_last_loop_time()) + sensor_struct->pitch_angle_filtered = ALPHA * (sensor_struct->pitch_angle_filtered + sensor_struct->theta_dot * loop_dt) + (1. - ALPHA) * accel_pitch; - sensor_struct->roll_angle_filtered = ALPHA * (sensor_struct->roll_angle_filtered + sensor_struct->phi_dot* get_last_loop_time()) + sensor_struct->roll_angle_filtered = ALPHA * (sensor_struct->roll_angle_filtered + sensor_struct->phi_dot* loop_dt) + (1. - ALPHA) * accel_roll; // Z-axis points upward, so negate distance //sensor_struct->lidar_altitude = -raw_sensor_struct->lidar_distance_m; - // Simply copy optical flow data + + //-------- Optical flow -----------// + // Copy over optical flow data sensor_struct->optical_flow = raw_sensor_struct->optical_flow; - flow_gyro_compensation(&sensor_struct->optical_flow, + flow_gyro_compensation(sensor_struct, raw_sensor_struct->lidar_distance_m, sensor_struct->roll_angle_filtered, sensor_struct->pitch_angle_filtered, - sensor_struct->currentQuadPosition.yaw); + sensor_struct->psi_dot); + //Filter OF velocities sensor_struct->optical_flow.xVelFilt = biquad_execute(&sensor_struct->flow_x_filt, sensor_struct->optical_flow.xVel); sensor_struct->optical_flow.yVelFilt = biquad_execute(&sensor_struct->flow_y_filt, sensor_struct->optical_flow.yVel); + + /* * Altitude double complementary filter */ diff --git a/quad/src/quad_app/type_def.h b/quad/src/quad_app/type_def.h index e1ffb3d4a0768bcaa05849efb2fb2ca5fb580cc0..e6e7d93f07ec966d525641cc1d75a759251fc3f1 100644 --- a/quad/src/quad_app/type_def.h +++ b/quad/src/quad_app/type_def.h @@ -328,6 +328,9 @@ typedef struct sensor { struct biquadState accel_z_filt; struct biquadState flow_x_filt; struct biquadState flow_y_filt; + struct biquadState phi_dot_filt; + struct biquadState psi_dot_filt; + struct biquadState theta_dot_filt; struct biquadState mag_x_filt; struct biquadState mag_y_filt; diff --git a/quad/xsdk_workspace/real_quad/src/hw_impl_zybo_optical_flow.c b/quad/xsdk_workspace/real_quad/src/hw_impl_zybo_optical_flow.c index 30d7a3381b677b54b0034810b72b58ba18619804..c45ca68a9c9063653b8eda70b4827c83e05fcea1 100644 --- a/quad/xsdk_workspace/real_quad/src/hw_impl_zybo_optical_flow.c +++ b/quad/xsdk_workspace/real_quad/src/hw_impl_zybo_optical_flow.c @@ -36,7 +36,7 @@ int zybo_optical_flow_read(struct OpticalFlowDriver *self, px4flow_t *of) { u8 buf[sizeof(i2c_integral_frame)]; // Read the sensor value - error = px4flow_read(i2c, buf, 0x16, sizeof(i2c_integral_frame)); + error = px4flow_read(i2c, buf, 0x16, 26); if(error == 0) { //Copy into struct @@ -47,7 +47,6 @@ int zybo_optical_flow_read(struct OpticalFlowDriver *self, px4flow_t *of) { of->quality = i2c_integral_frame.quality; of->dt = (double)i2c_integral_frame.integration_timespan / 1000000; } - return error; } @@ -66,7 +65,9 @@ int px4flow_read(struct I2CDriver *i2c, u8* recv_buffer, u8 register_addr, int s int error = 0; error = i2c->write(i2c, PX4FLOW_DEVICE_ADDR, buf, 1); - if (error) return error; + if (error) { + return error; + } error = i2c->read(i2c, PX4FLOW_DEVICE_ADDR, recv_buffer, size); return error; } diff --git a/quad/xsdk_workspace/real_quad/src/main.c b/quad/xsdk_workspace/real_quad/src/main.c index d11f783b9d9777390a7be79d1793acb7690ab525..f6991aca7623d2a99bdba5eb5a048e424874b5ff 100644 --- a/quad/xsdk_workspace/real_quad/src/main.c +++ b/quad/xsdk_workspace/real_quad/src/main.c @@ -49,7 +49,7 @@ int main() //test_zybo_mio7_led_and_system(); //test_zybo_i2c(); //test_zybo_i2c_imu(); - //test_zybo_i2c_px4flow(); + test_zybo_i2c_px4flow(); //test_zybo_i2c_lidar(); //test_zybo_i2c_all(); //test_zybo_rc_receiver();